]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
net/mlx4_en: Add support for destination MAC in steering rules
authorYan Burman <yanb@mellanox.com>
Wed, 12 Dec 2012 02:13:19 +0000 (02:13 +0000)
committerDavid S. Miller <davem@davemloft.net>
Wed, 12 Dec 2012 18:02:30 +0000 (13:02 -0500)
Implement destination MAC rule extension for L3/L4 rules in
flow steering. Usefull for vSwitch/macvlan configurations.

Signed-off-by: Yan Burman <yanb@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mellanox/mlx4/en_ethtool.c

index cc7bb25c7e40d79d3069a074a9cc4d3954c7702c..03447dad07e959082353ecd1408754736a61e2eb 100644 (file)
@@ -617,7 +617,13 @@ static int mlx4_en_validate_flow(struct net_device *dev,
        if (cmd->fs.location >= MAX_NUM_OF_FS_RULES)
                return -EINVAL;
 
-       switch (cmd->fs.flow_type & ~FLOW_EXT) {
+       if (cmd->fs.flow_type & FLOW_MAC_EXT) {
+               /* dest mac mask must be ff:ff:ff:ff:ff:ff */
+               if (!is_broadcast_ether_addr(cmd->fs.m_ext.h_dest))
+                       return -EINVAL;
+       }
+
+       switch (cmd->fs.flow_type & ~(FLOW_EXT | FLOW_MAC_EXT)) {
        case TCP_V4_FLOW:
        case UDP_V4_FLOW:
                if (cmd->fs.m_u.tcp_ip4_spec.tos)
@@ -745,7 +751,6 @@ static int mlx4_en_ethtool_to_net_trans_rule(struct net_device *dev,
                                             struct list_head *rule_list_h)
 {
        int err;
-       u64 mac;
        __be64 be_mac;
        struct ethhdr *eth_spec;
        struct mlx4_en_priv *priv = netdev_priv(dev);
@@ -760,12 +765,16 @@ static int mlx4_en_ethtool_to_net_trans_rule(struct net_device *dev,
        if (!spec_l2)
                return -ENOMEM;
 
-       mac = priv->mac & MLX4_MAC_MASK;
-       be_mac = cpu_to_be64(mac << 16);
+       if (cmd->fs.flow_type & FLOW_MAC_EXT) {
+               memcpy(&be_mac, cmd->fs.h_ext.h_dest, ETH_ALEN);
+       } else {
+               u64 mac = priv->mac & MLX4_MAC_MASK;
+               be_mac = cpu_to_be64(mac << 16);
+       }
 
        spec_l2->id = MLX4_NET_TRANS_RULE_ID_ETH;
        memcpy(spec_l2->eth.dst_mac_msk, &mac_msk, ETH_ALEN);
-       if ((cmd->fs.flow_type & ~FLOW_EXT) != ETHER_FLOW)
+       if ((cmd->fs.flow_type & ~(FLOW_EXT | FLOW_MAC_EXT)) != ETHER_FLOW)
                memcpy(spec_l2->eth.dst_mac, &be_mac, ETH_ALEN);
 
        if ((cmd->fs.flow_type & FLOW_EXT) && cmd->fs.m_ext.vlan_tci) {
@@ -775,7 +784,7 @@ static int mlx4_en_ethtool_to_net_trans_rule(struct net_device *dev,
 
        list_add_tail(&spec_l2->list, rule_list_h);
 
-       switch (cmd->fs.flow_type & ~FLOW_EXT) {
+       switch (cmd->fs.flow_type & ~(FLOW_EXT | FLOW_MAC_EXT)) {
        case ETHER_FLOW:
                eth_spec = &cmd->fs.h_u.ether_spec;
                memcpy(&spec_l2->eth.dst_mac, eth_spec->h_dest, ETH_ALEN);